Optimized String to_lower and to_upper:

[MisterPuma80]

Replaced slow binary tree search with simpler/faster hash map lookup. This also updates the ucaps_fetch.py script that generates the ucaps.h file.

Benchmark String to_lower/to_upper TLDR:

~27% faster small strings
~209% faster medium strings
~280% faster long strings

[MisterPuma80]

Benchmarks 4.5 before changes

String::to_lower short string, average ns: 42
String::to_lower medium string, average ns: 138
String::to_lower long, average ns: 830

String::to_upper short string, average ns: 40
String::to_upper medium string, average ns: 122
String::to_upper long string, average ns: 721

Benchmarks 4.5 with changes

String::to_lower short string, average ns: 33
String::to_lower medium string, average ns: 66
String::to_lower long string, average ns: 296

String::to_upper short string, average ns: 33
String::to_upper medium string, average ns: 63
String::to_upper long string, average ns: 270

[MisterPuma80]

Here is the benchmark script in a C++ GDExtention app.
example_gdextention.zip

[MisterPuma80]

When I want to run the benchmark, I essentially run a shellscript like this to build everything. Obviously this will depend on your setup.

# Download blazium
git clone https://github.com/blazium-engine/blazium
cd blazium
git checkout blazium-4.5
git submodule update --init --remote --recursive
# FIXME: Add any changes to the engine here ......
cd ..

# Download export templates
git clone https://github.com/blazium-engine/blazium-cpp

# Build engine and release templates
cd blazium
scons platform=linuxbsd target=editor dev_build=no dev_mode=no use_llvm=yes linker=mold tests=yes execinfo=yes scu_build=yes dotnet=yes -j 25
scons platform=linuxbsd target=template_release dev_build=no dev_mode=no use_llvm=yes linker=mold scu_build=yes -j 25
cd ..

# Dump fresh API json file and header from Blazium Editor
cd blazium-cpp/gdextension
rm -f extension_api.json
rm -f gdextension_interface.h
../../blazium/bin/blazium.linuxbsd.editor.x86_64.llvm --headless --dump-extension-api --dump-gdextension-interface
cd ../..

# Build benchmark extention
scons platform=linux target=template_release dev_build=no use_llvm=yes -j 25
./blazium/bin/blazium.linuxbsd.editor.x86_64.llvm --headless --export-release "Benchmarks Headless Linux" ./demo/project.godot

# Run benchmark
./demo/export/demo.x86_64 --no-header --headless --single-threaded-scene --fixed-fps 60 --delta-smoothing disable --disable-vsync --disable-render-loop

[jss2a98aj]

I wonder if bit manipulation on certain ranges might be any faster. Regardless, will run some benchmarks on my system after I do the next round of cherry-picks.

[MisterPuma80]

I bet we can make it way faster, but I didn't want to change the structure too much from upstream. Not yet anyway.